Table 7.

Comparison of pose estimation algorithms based on deep learning.

Methods	Level	Advantages or Applicable Scenarios	Limitation
Regression-based methods	Instance-level	Simple design and wide application.	Applicability to complex environments may be limited.
Feature-based methods	Instance-level	Situations with rich features and not severe occlusion.	Symmetry needs to be considered.
Fusion-based methods	Instance-level	Industrial applications, are suitable for occlusion.	The method design is relatively complex.
Point cloud-based methods	Instance-level	Robot grabbing-related tasks.	Surface reflections may result in poorer results.
Regression-based methods	Category-level	Everyday objects, perform better in generalization.	Poor handling of intra-category differences.
Prior-based methods	Category-level	More robust to intra-class differences and color changes.	High demand for computing resources.